fix(CT-GF1): review fix round — literal DrawHere clip shape, empty-clip cull, popup input routing

Applies all 11 items from the Opus dual-lens review of 989f6652 (0
blockers, 7 SHOULD-FIX, 4 NOTE):

- S2: UiElement.DrawSelfAndChildren now pushes the ambient clip right
  after PushAlpha and wraps OnDraw + the children walk +
  OnDrawAfterChildren in ONE block — the literal UIRegion::DrawHere
  @0x0069FA30 shape, which clips an element's OWN DrawSelf too, not
  just its children (UIElement_Text::DrawSelf @0x00467AA0 locks glyph
  blits to its own clipped surface rect; UIRegion::DrawSelf
  @0x0069F1A0 blits per clip rect). Deleted the two now-redundant
  ad-hoc self-clips this supersedes: UiText.DrawText and
  UiField.DrawMultiLine both pushed their own (0,0,Width,Height) —
  exactly what the new ambient clip already provides one level up.
  Kept UiButton.DrawBlockLabel's clip: it clips to LabelBox/ValueBox,
  an authored INNER sub-rect that can be smaller than and offset from
  the button's own full rect — a genuine narrower viewport, not a
  redundant duplicate.
- S3: deleted UiItemList's `ClipsChildren => CellWidth > 0f` override
  — correct under the old opt-in-false default, inverted under the
  new default-true (an unconfigured list would stop clipping instead
  of clipping like everything else).
- S4: pinned the escaped-popup input path end to end. New
  UiAncestorClipTests test mounts a menu inside a short window on a
  real UiRoot, opens it, and proves a click in the escaped popup
  region reaches the menu through UiRoot.PopupHit (a plain top-down
  walk is proven to reject the same point first). UiRoot.WantsMouse
  now also checks PopupHit — it previously only checked Captured/
  HitTestTopDown, so a game action could fire underneath an open
  dropdown's escaped region. OnMouseDown/OnScroll already routed
  through PopupHit first (#374); unchanged.
- S5: strengthened the Titles-divider regression test's positive
  half. The old assertion only checked SOME quad's Y fell in a band —
  vacuously true given other same-band content. Now asserts the
  divider's exact rect (X and Y), then diffs against the same rect
  with the divider hidden (Visible=false) to prove the quad was
  actually attributable to it.
- S1: added UiWindowDrawCaptureSweepTests — Character/Chat/Vendor/
  Options mounted through their real production Bind entry points
  with a non-zero sprite resolver, drawn via RecordingGpuDevice,
  asserting a per-window vertex floor (~40-45% of this session's
  observed baseline: Character 588, Chat 162, Vendor 54, Options 240)
  plus one key sprite id read LIVE off the bound controller/element
  (never hardcoded). Character's key sprite (RetailChromeSprites.
  TopEdge) specifically exercises OnDrawAfterChildren, the exact path
  S2's caution note flagged. Inventory/Paperdoll/social/map-house
  skipped — no single fixture-driven top-level Bind entry point.
- S6: added the CT-GF1 subsection to the campaign plan's ledger
  (989f6652 + this fix round; CT7 re-gate still owed).
- S7: UiRenderContext.PushClipUnbounded now resets to the CANVAS rect
  (0,0,ScreenSize), not null — retail's own popup region is
  SCREEN-clipped (UIElement_Menu::MakePopup spawns a top-level region
  bounded by the screen), not truly unbounded. AD-113 amended.
- N1: UiRoot overrides ClipsChildren => false — the root's own region
  IS the screen (the viewport already scissors it), so this is a
  safety net against a momentarily zero-sized root silently blanking
  the whole UI tree under the new ancestor-clip default.
- N2: added the empty-clip subtree cull (retail's var_24 gate
  @0x0069FB8E) to DrawSelfAndChildren only — DrawOverlays is a wholly
  separate traversal untouched by this change. New test proves a menu
  inside a fully-clipped (zero-width) window still draws its open
  popup via the overlay pass while the main pass draws nothing.
- N3: CT7 script §5 now names the collapsed-toolbar check and the
  four highest-overflow windows (combat/vitals bar, Options
  bottom-button row, map/house page, floaty chat) as explicit
  eyeball items for the re-gate.
- N4: verification below covers both the working tree and the clean
  committed tree.

Decomp anchors: UIRegion::DrawHere @0x0069FA30 (var_24 gate
@0x0069FB8E); UIElement_Text::DrawSelf @0x00467AA0 (self-clip);
UIRegion::DrawSelf @0x0069F1A0; UIElement_Menu::MakePopup (screen-
clipped popup region).

Verification (both runs green, --filter "Lane!=InstalledDat&
Lane!=PreparedPackage&Lane!=Live&Lane!=Manual&Lane!=Timing&
Lane!=Windows&Lane!=Linux&Lane!=SystemFont&Purpose!=Diagnostic&
Status!=KnownFailure"): full Release solution build green; working
tree 14,900+ tests across every project (one LandblockPresentation
PipelineTests flake reproduced ONLY under full-solution parallel
load, passes standalone and on rerun — unrelated to this change,
streaming domain); InstalledDat lane green (ACDREAM_RUN_INSTALLED_DAT
_TESTS=1, Status!=KnownFailure, 205+34+3+172 App/Content/Bake/Core
tests). Clean committed tree (git stash push -u the uncommitted
owner probe + docs files, rerun, stash pop) reported in the session
summary.

src/AcDream.App/UI/UiRoot.cs carries an unrelated, pre-existing
uncommitted owner probe (ACDREAM_PROBE_UI_HOVER) — staged selectively
(git add -p) so only this commit's own two hunks (ClipsChildren
override, WantsMouse) landed; the probe hunk is untouched and stays
uncommitted, same as before this fix round.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
This commit is contained in:
Erik 2026-08-25 07:11:42 +02:00
parent 989f665214
commit 025108a8aa
13 changed files with 717 additions and 120 deletions

View file

@ -33,6 +33,19 @@ public sealed class UiRoot : UiElement
/// <summary>Single owner for named retained-window lifecycle and raise policy.</summary>
public RetailWindowManager WindowManager { get; }
/// <summary>
/// CT-GF1 fix round (N1): the root's own region IS the screen — the viewport
/// itself already scissors everything drawn to it, so retail has no analog of
/// clipping the root to its OWN tracked (Width,Height) the way
/// <see cref="UiElement.ClipsChildren"/>'s new client-wide default would.
/// Overriding false here is a safety net, not a cosmetic choice: without it, a
/// root momentarily reporting a zero (or stale, pre-first-resize) size would
/// silently blank the ENTIRE UI tree — every top-level window culled by the new
/// ancestor-clip default's empty-intersection gate — rather than the intended
/// "root passes its full extent through to its children uninterpreted."
/// </summary>
protected override bool ClipsChildren => false;
/// <summary>
/// Campaign LA gate round 2 (register AD-98): when set, the retained tree
/// is laid out in this fixed authored canvas (the char-select screen's
@ -189,8 +202,27 @@ public sealed class UiRoot : UiElement
/// The host ORs this into the InputDispatcher's WantCaptureMouse gate so game
/// actions (movement, world-pick) are suppressed while the user interacts with
/// a retail window — mirrors ImGui's WantCaptureMouse.
///
/// <para>
/// CT-GF1 fix round (S4): also checks <see cref="PopupHit"/> — an open UiMenu
/// dropdown's escaped region (the part of the popup that extends outside its
/// owning window's own ancestor-clipped bounds, e.g. a channel menu opened
/// upward past a short chat window's top edge) is reachable by
/// <see cref="OnMouseDown"/>/<see cref="OnScroll"/> through <see cref="PopupHit"/>
/// specifically BECAUSE it bypasses the ordinary top-down <see cref="HitTest"/>
/// walk's ancestor clip gate (CT-GF1's new client-wide
/// <see cref="UiElement.ClipsChildren"/> default would otherwise reject that
/// point before ever reaching the popup's own out-of-bounds <c>OnHitTest</c>
/// union — see #374's popup routing doc above). Without this,
/// <see cref="HitTestTopDown"/> alone would report "no widget here" for a point
/// the pointer visibly sits over, letting a world click/movement action slip
/// through underneath the open popup.
/// </para>
/// </summary>
public bool WantsMouse => Captured is not null || HitTestTopDown(MouseX, MouseY).element is not null;
public bool WantsMouse =>
Captured is not null
|| PopupHit(MouseX, MouseY) is not null
|| HitTestTopDown(MouseX, MouseY).element is not null;
/// <summary>True when a widget holds keyboard focus (e.g. a focused chat input).</summary>
public bool WantsKeyboard => KeyboardFocus is not null;
@ -1386,6 +1418,23 @@ public sealed class UiRoot : UiElement
Data1: (int)(x - screen.X),
Data2: (int)(y - screen.Y));
w.OnEvent(in enter);
if (UiDiagnostics.ProbeHover)
{
string detail = w is UiScrollbar sb
? $" start=0x{sb.ActiveStartSpriteForTest:X8}"
+ $" end=0x{sb.ActiveEndSpriteForTest:X8}"
+ $" thumb=0x{sb.ActiveThumbSpriteForTest:X8}"
+ $" disabled={sb.IsModelDisabled}"
: "";
Console.WriteLine(
$"[ui-hover] widget={w.GetType().Name}"
+ $" dat=0x{w.DatElementId:X8}"
+ $" local=({(int)(x - screen.X)},{(int)(y - screen.Y)}){detail}");
}
}
else if (UiDiagnostics.ProbeHover)
{
Console.WriteLine("[ui-hover] widget=<none>");
}
}